<![CDATA[Modern robotic systems address complex engineering challenges using artificial intelligence and machine learning techniques. In agricultural robotics, fruit identification and sorting remain challenging due to variations in size, shape, color, orientation, and lighting conditions. This study presents the design and implementation of a vision-based fruit sorting robotic system integrating YOLOv8-based object detection with robotic manipulation. A custom dataset consisting of images of 2 different fruits (namely banana and strawberry images), including single-fruit and multi-fruit scenarios, was used and manually annotated using bounding boxes in CVAT. The dataset was divided into training, validation, and test subsets to enable robust model development under realistic operational conditions. A lightweight YOLOv8 model was trained using CUDA acceleration and optimized for edge deployment by selecting YOLOv8n to balance inference speed and detection accuracy. The trained model was converted to ONNX format and deployed on a Raspberry Pi 5 for real-time inference using live camera input. Evaluation on an independent test dataset achieved a precision of 0.999, recall of 1.000, mAP@0.5 of 0.995, and mAP@0.5:0.95 of 0.963 under controlled experimental conditions with limited object classes. The modular architecture enables low-cost and scalable deployment and provides a foundation for future enhancements, including closed-loop robotic control, additional object categories, and operation in more dynamic environments.]]>
Copyrights © 2026
